Animacy, Generalized Semantic Roles, and Differential Object Marking

Animacy, Generalized Semantic Roles, and
Differential Object Marking
Beatrice Primus, University of Cologne
Abstract This chapter addresses the role of case and animacy as interacting cues
to role-semantic interpretation in grammar and language processing. Animacy is
interpreted as a cue to agentivity taken as a multidimensional, generalized
semantic role. In this view, several agentivity properties entail or strongly
correlate with animacy on the part of the respective participant. In contrast, none
of the patient characteristics presuppose an animate participant. By abductive
reasoning animacy is used as a probabilistic cue to agentivity. The empirical focus
of this chapter lies on animacy-driven differential object marking (DOM). The
selection of the case marker in the DOM-patterns under consideration, which was
assumed to be triggered by the animacy of the second argument in previous
approaches, is explicable by role-semantic constraints tied to agentivity. This view
explains some DOM-related phenomena that remain unexplained in other
approaches. The close connection between animacy and role-semantic
interpretation is also manifest in language processing. This chapter reports
experimental studies showing that the brain areas and the neuronal patterns that
react to animacy effects are also involved in the interpretation of semantic roles.
On a more general level, taking animacy as a cue to agentivity contributes towards
a better understanding of the basic notions that characterize agentivity.
1. INTRODUCTION
This chapter 1 explores the connection between animacy and semantic roles in
patterns of case selection where the animacy of the noun phrase referent is
assumed to trigger differential object marking (DOM) in previous approaches. In
this traditional view, which will be criticized below, case marking is considered to
be a cue to animacy. The basic DOM-patterns and some facts that are still
unexplained in alternative approaches will be presented in section 2 of this
chapter.
1 I thank Marco García García for his help with the Spanish data, Thomas Anzenhofer and Jana Koshy
for their help with the Malayalam data, and the editors of this volume and an anonymous reviewer for
their comments on a previous version of this chapter.

2
By using generalized semantic roles, i.e. decomposing a few general roles into
more basic notions, I take a different view on the DOM-patterns under
consideration. The interaction between animacy and semantic roles turns out to be
closer than usually assumed. Many agentive properties – volition (or control),
sentience, alienable possession, and certain patterns of motion – imply the
involvement of a higher animate participant. By contrast, no patient-like concept
implies animacy on the part of the respective participant. This will be shown in
section 3 of this chapter. This type of approach to semantic roles offers a
straightforward explanation for phenomena connected to DOM that remain
unexplained in previous approaches. Taking case in DOM-patterns as a cue to
animacy, as in previous approaches, fails to explain why the same case may be
used for certain inanimate patients, why it may be missing for animate patients
with ditransitive verbs, why it may be conditioned by the semantic verb type or by
individual verb lexemes, and finally, why it is used as a cue to semantic roles in
other contexts of use (section 4). I will show for Spanish that case selection in
animacy-driven DOM is explicable by role-semantic case constraints tied to
agentivity (section 5). The present approach takes a role-functional view on
animacy for the phenomena under discussion: animacy is only relevant as a cue to
role-semantic interpretation. This view does not eliminate animacy from linguistic
explanation. On the contrary, a better understanding of the basic notions that
characterize semantic roles may be achieved by taking animacy into account
(section 6).
2. ANIMACY-BASED DIFFERENTIAL OBJECT MARKING
In some languages certain direct objects are marked by an object marker, while
other objects remain unmarked. In the traditional view, this pattern of variation is
assumed to be driven by semantic features of the noun phrase such as animacy,
definiteness or specificity, or a combination of these factors. I will focus on
animacy-based variation. Definiteness and specificity will be addressed later
(sections 3 and 6). I will illustrate DOM by examples from Spanish, an accusative
language, in (1a, b) and from Hindi, an ergative language, in (2a, b). These are
two languages that have been considered to have a typical DOM-pattern.
Spanish (García García 2007: 63)
(1) a. Conozco *este actor / a este actor.
know:PRS.1SG this:M.SG actor / OBJ this:M.SG actor
‘I know this actor.’
b. Conozco esta película / *a esta película.
know:PRS.1SG this:F.SG film / OBJ this:F.SG film
‘I know this film.’
Hindi (Mohanan 1994: 80)

3
(2) a. Ilaa-ne bacce-ko / *baccaa ut h aayaa.
Ila-ERG child-OBJ / child lift.PRF
‘Ila lifted the / a child.’
b. Ilaa-ne haar-ko / haar ut h aayaa.
Ila-ERG necklace-OBJ / necklace lift.PRF
‘Ila lifted the / a necklace.’
In Standard European Spanish, which is discussed in this chapter, animacy seems
to be the most important factor. (1b) shows that a definite or specific object is not
marked by the differential object marker, the preposition a, unless the referent is
human or at least animate, as in (1a). In Hindi, definiteness plays a more important
role. The differential object marker, the postposition ko, must be used with human
noun-phrase referents, which can be interpreted as definite or indefinite, as shown
in (2a). If an inanimate participant is marked, it is interpreted as definite, if it is
unmarked, it is indefinite, as shown in (2b). 2 In both languages, animacy, more
specifically a human participant in Hindi, seems to be a sufficient condition for
overt marking. With indefinite or unspecific noun phrases, animacy (or being
human) is also a necessary condition for DOM. 3 In Hindi and other Indic
languages, the ergative marker is only used in the perfect. In other tenses the
respective argument is zero-marked. This leads to case ambiguity if the patient is
zero-marked as well.
The animacy criterion turns out to be less reliable than usually assumed,
if one takes more data into consideration. In Spanish, animacy is not a necessary
condition, as DOM occurs with certain inanimate, as shown in (3):
Spanish (García García 2007: 64)
(3) En esta receta, la leche puede sustituir al huevo.
in recipe the:F.SG milk can:PRS.3SG replace OBJ:DEF.M.SG egg
this:F.SG
‘In this recipe, egg can be replaced by milk.’
As we will see later (cf. section 4), the verbs showing DOM with inanimate form a
uniform group in terms of role semantics. This indicates that role semantics
strongly influences DOM.
Furthermore, animacy is not a sufficient condition in Spanish, as there is no
DOM with animate patients of ditransitive verbs, as shown in (4a, b):
2 In ergative languages, the zero-marked patient can be interpreted either as a direct object or as a
morphosyntactic subject, depending on the language and the phenomenon under discussion (cf. Dixon
1994, Primus 1999). The zero-marked patient will be treated as an object in this chapter, for
convenience only, following commonly used terminology.
3 These observations are rough estimates. Acceptability judgements vary considerably in some cases.

4
(4) a. La madre le da *al niño al padre.
the:F.SG mother CLT.3SG.DAT give:PRS.3SG OBJ:DEF.M.SG child OBJ:DEF.M.SG father
‘Mother gives the child to the father.’
b. La madre le da el niño al padre.
the:F.SG mother CLT.3SG.DAT give:PRS.3SG the:M.SG child OBJ:DEF.M.SG father
‘Mother gives the child to the father.’
The lack of DOM in ditransitives in Spanish (and in some other languages with
DOM) is explicable by the fact that the animate object marker is identical with the
recipient marker, i.e. the dative. The recipient takes the dative leaving the animate
patient unmarked, as double datives are prohibited in Standard European Spanish.
This is another type of construction in which role selection (the recipient)
influences DOM.
In some languages, DOM is extended to ditransitive constructions with two
animate objects (cf. Kittilä 2006). This is illustrated in (5) by examples from
Punjabi:
Punjabi (Bhatia 1993: 89)
(5) a. mãi mãã nüü kaake nüü dittaa.
I mother DAT child DAT give.PST
‘I gave the child to the mother.’
b. mãi mãã nüü kaake dittaa.
I mother DAT child give.PST
‘I gave the child to the mother.’
c. *mãi mãã kaake nüü dittaa.
I mother child DAT give.PST
‘I gave the child to the mother.’
d. mãi kuRii nüü vekhiaa.
I girl DAT see.PST
‘I saw a girl.’
If a ditransitive sentence has a recipient object and an animate patient object, then
both objects can receive the dative marker nüü in some dialects of Punjabi, as
shown in (5a). 4 An animate object of a monotransitive clause receives the same
4 In Kittilä’s (2006) cross-linguistic survey, the animate patient usually inherits the marker of the
recipient. Languages with DOM in ditransitives vary with respect to recipient-marking. In some
languages, the recipient retains its marking yielding double dative constructions, as shown in the
Punjabi example (5a). In other languages the recipient looses its canonical marker in favour of the
animate patient and takes another oblique marker (e.g. Sahaptin).

marker, cf. (5d). This means that the DOM pattern of monotransitive clauses is
extended to ditransitive clauses as well. The ambiguity resulting from the double
dative marking is resolved by word order (Bhatia 1993: 89): the recipient has to
precede the direct object.
In the Majhi dialect of Punjabi it is only the recipient that takes the dative, as
shown in (5b) and (5c). This pattern was also illustrated with Spanish examples in
(4) above. A similar situation is found in Hindi (cf. Mohanan 1994: 85).
The fact that DOM is conditioned by the semantic verb type or by individual
verb lexemes is another indicator that animacy and role subcategorization are
interrelated. Hindi is illustrative in this respect (cf. Mohanan 1994: 81f.). The
objects of the verbs lik h ‘write’, banaa ‘make’, pad h ‘read’, gaa ‘sing’, and pii
‘drink’, which subcategorize for inanimate objects in Hindi, do not allow the
postposition ko even if the object is definite. Conversely, verbs that subcategorize
for animate objects such as piit ‘beat, spank’, maar ‘kill’, and bulaa ‘call’ tolerate
only the postposition ko. Thus, a verb may impose an animacy restriction on its
object, suspending the other conditions for alternation.
Likewise, the diachronic development of DOM in Spanish is conditioned by
the semantic class of the verb (cf. von Heusinger 2008). Verbs with a high
preference for human objects such as matar ‘kill’ and herir ‘hurt’ were the first to
acquire DOM with full noun phrases.
Another type of evidence suggesting that role semantics and animacy are
closely related is the choice of the animacy marker. This marker encodes semantic
roles in other contexts of use (cf. Bossong 1985, 1998, Lazard 2001). In most of
the Indic, Iranian, Semitic, Tupi-Guarani, and Romance languages, the animacy
marker is the same as that used for the recipient, addressee, and benefactive (i.e.
the dative) in ditransitive constructions, as illustrated above. In many Romance,
Iranian and Indic languages the animate patient marker also codes the experiencer
of certain psychological predicates. In some Iranian languages (e.g. Jaghnobi) it
indicates the agent of transitive clauses (i.e. the ergative). In sum, animacy
markers and role-semantic markers are formally related in a systematic way. The
semantic roles that donate their marker to the animacy-based function are
restricted to roles that include agentive components in approaches using
generalized semantic roles, as will be shown in the next section. Such roles are
recipients, addressees, benefactives, and experiencers.
The close connection between role-semantic interpretation and animacy is
also manifest in language processing. Neurolinguistic research has established that
the brain areas and the neuronal patterns that react to animacy effects are also involved
in the interpretation of semantic roles. Research using event-related potentials
has found the N400 component, a central-parietal negativity at approximately
400 ms post onset of the verbal stimulus, whenever the subcategorization frame of
a head is violated (cf. Friederici 1999). This component occurs, for instance, with
role-semantic interpretation problems and with certain case violations in German.
In addition, N400 was elicited as a reaction to certain animacy effects (cf. Frisch
and Schlesewsky 2001, Schlesewsky and Bornkessel 2004, for German; Bornkessel-Schlesewsky
and Schlesewsky 2009, for other languages, Wang et al., this
volume, for Chinese). Pertinent German data are shown in (6):
5

6
(6) a. welchen Mönch der Bischof begleitete
which:ACC monk the:NOM bishop accompany:PRT.3SG
‘which monk the bishop accompanied’
b. welchen Mönch der Zweig streifte
which:ACC monk the:NOM twig brush:PRT.3SG
‘which monk the twig brushed’
c. welcher Mönch der Bischof begleitete
which:NOM monk the:NOM bishop accompany:PRT.3SG
‘which monk the bishop accompanied’
d. welcher Mönch der Zweig streifte
which:NOM monk the:NOM twig brush:PRT.3SG
‘which monk the twig brushed’
The ungrammatical condition with two animate, identically case marked
arguments (6c) triggered a N400 at the point of the second argument in
comparison to the control condition (6a). Bornkessel, Frisch, and Schlesewsky
interpret their findings as follows. N400 in (6c) reflects the inability to hierarchize
the arguments with respect to one another in terms of semantic roles. By contrast,
in sentences with two identically case-marked arguments that differ in animacy, as
in (6d), the language processing system takes animacy as a cue for a role-semantic
hierarchization of the arguments. Hence, there is no N400 effect in (6d) in
comparison to (6b). A parietal positivity (P600), which occurs in both
ungrammatical conditions, indicates a case reanalysis of the second NP.
Neuroimaging research also indicates that animacy and role-semantic
interpretation are closely connected. Grewe and colleagues (Grewe 2006, Grewe
et al. 2005, 2006, 2007) have observed that deviations from co-argument
distinctness in terms of animacy, as in (7b), correlate with activation in the
posterior portion of the left superior temporal sulcus (pSTS) in comparison to the
control sentence (7a), in which the co-arguments show a canonical animacy
asymmetry. This effect is independent of the relative word order of the coarguments.
(7) a. Wahrscheinlich hat der Mann den Garten gepflegt.
probably have:PRS.3SG the:NOM man the:ACC garden takecare.PART.PRF
‘The man has probably taken care of the garden.’
b. Wahrscheinlich hat der Mann den Direktor gepflegt.
probably have:PRS.3SG the:NOM man the:ACC director takecare.PART.PRF
‘The man has probably taken care of the director.’

7
Activation in this brain area was also observed in studies investigating the neural
correlates of agency and motion patterns that are attributed to animates (cf.
Pelphrey et al. 2004, Schultz et al. 2004, Schultz et al. 2005, Thompson et al.
2005). Activation in the pSTS, as observed for (7b), is not likely a lexical
phenomenon due to the intrinsic animacy of the noun phrase referent. Animate vs.
inanimate nouns at the word (or picture) level lead to activation differences in
regions other than the left pSTS (cf. Bornkessel-Schlesewsky and Schlesewsky
2009). In sum, neuroimaging studies also support the view that animacy and rolesemantic
interpretation are closely connected.
Another pertinent study shows that animacy asymmetries in co-ordinated
noun phrases do not trigger reliable effects when the animate-before-inanimate
constraint is violated in language production (cf. Branigan et al. 2008). Our
interpretation of this finding is that the semantic roles of co-ordinated noun
phrases are identical. As a result, animacy asymmetries in co-ordinated noun
phrases cannot be used to retrieve role-semantic information.
In sum, there is both grammatical and neurolinguistic evidence that animacy
is closely connected to sentence interpretation in terms of semantic roles. Patterns
of case variation that are assumed to be determined by animacy in previous
approaches may occur with inanimate objects and are absent with animate objects
in ditransitive constructions in some languages. The semantic class of the verb
influences this type of case variation. The case marker found in DOM is also used
for semantic roles that show an affinity to agents. Neurolinguistic research has
established that the brain areas and the neural patterns that react to animacy
violations are also involved in the interpretation of semantic roles. When semantic
role distinctions are absent, as in co-ordinated noun phrases, animacy ceases to
influence noun phrase word order. In order to reveal the close interaction between
animacy and semantic roles we need an approach that uses generalized semantic
roles.
3. GENERALIZED SEMANTIC ROLES AND ANIMACY
ENTAILMENTS
Earlier work on semantic roles, notably Fillmore (1968), used animacy as a
characteristic property of several roles, deep cases in his terminology. For him, the
agentive is the typically animate instigator of the action, the dative is characterized
as the animate being affected by the state or action, while the instrumental is the
inanimate force or object causally involved in the action or state (1968: 24). The
main critical argument against Fillmore’s view was that animacy is a categorial
semantic distinction, an intrinsic (or inherent) property of a noun phrase referent,
whereas semantic roles are relational concepts determined by the semantic
subcategorization frame of a predicate. In reaction to this criticism, Fillmore
(1977) and subsequent research on semantic roles have discarded animacy as a

8
role-semantic criterion. Unsurprisingly, correlations between animacy and certain
semantic roles, including agent and recipient (Fillmore’s dative), are often noted
only in passing (cf. VanValin and LaPolla 1997, de Swart et al. 2008).
The interaction between animacy and semantic roles turns out to be closer
than usually assumed in approaches that decompose semantic roles into more
basic notions (cf. Lakoff 1977, Dowty 1991, Primus 1999, Ackerman and Moore
2001). Semantic decomposition reduces the inventory of superordinate,
generalized roles dramatically without neglecting finer-grained distinctions.
Dowty’s account, which needs only proto-agent and proto-patient as generalized
roles, will be taken as a theoretical starting point in this chapter. In this kind of
approach, nearly all agentive properties but none of the patient-like properties
entail the involvement of a higher animate participant, as will be shown below.
The agent proto-role is characterized by Dowty (1991: 571-572) and Primus
(1999, chap. 4) as follows:
(8) Proto-agent entailments:
a. x does a volitional act: John refrains from smoking.
b. x is sentient of or perceives another participant: John knows / sees /
fears Mary.
c. x causes an event or change of state in another participant: His
loneliness causes his unhappiness.
d. x is moving autonomously: Water filled the boat.
e. x is a possessor of another entity: Peter has a new car.
Although most verbs select more than one proto-agent property for their subject
argument (e.g. murder, nominate, or give), each of these properties can occur in
isolation as shown by the subject argument in the examples in (8a) – (8e). The
properties mentioned in (8a) – (8d) are Dowty’s proposal (1991: 572). 5 Possession
is included as an agentive component following, among others, Jackendoff (1990).
Each of these characteristics is semantically independent. Nevertheless, some of
them tend to co-occur (e.g. volition or causation and movement) and one property
may unilaterally imply another. Thus, for instance, volition implies sentience (cf.
Dowty 1991: 606).
The concepts defining proto-agent are nothing new to the linguistic
community: volition or control, causation, movement, sentience, and possession.
Volition is used by Dowty in the sense of intentionality: the participant in question
intends this to be the kind of act named by the verb. In von Wright’s (1963)
approach intentionality is a teleological causal concept that defines the notion of
action.
Sentience is used as a cover term for a cognitive state, emotion, or perception.
Including sentience in the list of proto-agent properties is uncommon within
linguistic tradition, but it is in conformity with neuroscientific research (Ochsner
5 Tentatively Dowty also mentions independent existence as a proto-agent entailment. This means that
the participant exists independently from the situation denoted by the verb. This property is omitted
here because it is a reference-related property that should be treated separately.

and Gross 2005, Gross 2007). This line of research assumes that a particular
mental state is produced by the experiencer in reaction to an external stimulus and
that emotions, in particular, can be cognitively controlled by the experiencer.
Movement is a proto-agent property only if it is autonomous, i.e. not caused
by another participant (cf. Dowty 1991: 574). This is in conformity with cognitive
linguistic research that demonstrates the relevance of the concept of self-propelled
movement for the cognitive development of the notion of agentivity and causation
(cf. Leslie 1995, Premack and Premack 1995). If movement is caused by another
participant, it will be considered a proto-patient property in the present approach.
Thus, for instance, in John threw the ball both entities move, but only John, the
proto-agent, moves in a self-induced way. The ball, the proto-patient, moves as a
response to John’s movement.
As to possession, Premack and Premack (1995: 193f.) point to a crucial
difference between the notion of group and that of possession. Both notions imply
that two or more objects are physically or spatially connected and capable of comovement.
But only possession is characterized by the ability of the possessor to
control the possessed object. Alienable possession has this agentive quality, as
shown by Folli and Harley (2008). Thus, for instance, Peter has three cars, is
perfectly acceptable as opposed to *The garage has three cars in a possessive
reading. With inalienable possession, potential control and the correlating animacy
contrast are irrelevant: Peter has black hair. The garage has a front window.
The most important departure of the present approach from Dowty is the
central status given to the various manifestations of causation, including
psychological concepts related to volition and control, and to the distinction
between independent and dependent involvement (cf. Primus 1999, 2006, Lamers,
this volume). In this view, a general role-semantic dependency notion is the
underlying criterion that distinguishes proto-agents from proto-patients. The
patient proto-role is defined in this approach as follows: its kind of involvement is
dependent on the kind of involvement of another participant, the proto-agent.
Consequently, the list of properties characterizing the proto-patient is derived
from the basic notions in the definition of the proto-agent. Thus, for instance,
causal affectedness as a proto-patient property is the converse of the causer notion
in the proto-agent list. In this approach, the converse relation is generalized over
all involvement properties: controller vs. controlled, causer vs. causally affected
(“changed”), mover vs. moved, experiencer vs. experienced (“stimulus”), and
possessor vs. possessed. This view on proto-patient departs from Dowty’s
approach. Dowty’s proto-patient entailments are change of state, incrementally
affected (i.e., incremental theme), causally affected, and either stationary or
moving as a result of being causally affected. The two views on proto-patient have
in common that none of the proto-patient properties entail the involvement of an
animate participant. The co-argument dependency model of Primus (1999, 2006)
is better suited to incorporate different types of co-argument dependencies.
Pertinent to DOM is the connection between role-semantic and reference-related
co-argument dependencies that will be discussed further below in this section.
Furthermore, this model is in line with the neurolinguistic Argument Dependency
Model of Schlesewsky and Bornkessel (2004, cf. also Wang et al., this volume).
9

10
A specific trait of the generalized role approach that makes it particularly
suitable to explain DOM is that it allows for arguments having a combination of
proto-agent and proto-patient properties. Independently of DOM, roles such as
recipient, addressee or benefactive are mixed. They are selected by verbs denoting
a transfer in possession (x gave y something z, x baked y a cake z) or a caused
sentience (x told y a story z, x showed y a picture z). As a possessor or an
experiencer of the entity z, the argument y in these examples is a proto-agent
relative to the entity z. At the same time it is a proto-patient relative to the first
argument x, which causes the situation denoted by the verb and causally affects y.
Animacy-driven DOM involves a similar role overlap. It occurs whenever a noun
phrase subcategorized for patient properties acquires potential agent properties
due to its intrinsic meaning (cf. (10) in section 4 of this chapter).
In sum, what characterizes agentive participants is that they are capable of
producing the activity or state described by the predicate by themselves. The
capacity to control the situation or other entities involved in the situation captures
a common trait of several agentive notions that were subsumed under volition,
sentience and alienable possession. These three agentive properties entail the
involvement of a higher animate participant. By contrast, none of the patient-like
properties entails animacy on the part of the respective participant, no matter what
type of approach to the patient-role one favours (e.g. Dowty 1991 or Primus 1999,
2006).
As to the agentive notion of autonomous movement, a certain type of
autonomous motion implies an animate participant. A moving object – a spot of
light or a geometric figure in the pertinent neuroscientific experiments – can create
the subjective impression that it is alive, based on its pattern of movement, when
this pattern suggests that it is caused by an internal rather than an external source
of energy (cf. Gelman et al. 1995, Tremoulet and Feldman 2000, Scholl and
Tremoulet 2000, Opfer 2002). As mentioned above, neuroimaging studies also
confirm the close neural connection between the detection of agency and the
attribution of certain motion patterns to animates (cf. Pelphrey et al. 2004, Schultz
et al. 2004, Schultz et al. 2005, Thompson et al. 2005). A remarkable result of
these experimental studies is that they report an impression of animacy despite the
fact that the entity presented in the experiments had no resemblance to an animate
being except for its abstract movement pattern. This result supports the
assumption defended here that animacy has to be regarded in role-functional
terms: anything that has an agentive behaviour that is associated with higher
animate beings, i.e. volition, a ‘biological’ motion pattern, sentience or alienable
possession, is treated cognitively as a higher animate entity.
The co-argument dependency model defended here contributes to a better
understanding of the connection between animacy-based and reference-related
DOM. The basic assumption is that a clear-cut asymmetric co-argument
dependency is preferred in grammar and language processing. This means in rolesemantic
terms that one argument has only agentive properties and the other
argument only patient-properties. I will argue that in both types of DOM, an
animate or definite second argument blurs maximal distinctness in terms of coargument
dependency. Recall that in Hindi, for instance, an animate argument or a

11
definite (possibly inanimate) argument is overtly case-marked (cf. example (2b) in
section 2). As mentioned above, in my view, the patient proto-role is defined by
its co-argument dependent status relative to the proto-agent. Under this
assumption, the role hierarchy proto-agent > proto-patient is based on an
asymmetric co-argument dependency. In animacy-based DOM, the animacy of the
proto-patient leads to a situation in which asymmetric co-argument dependency
cannot be established unequivocally.
Reference-related distinctions, i.e. definite > indefinite or specific > nonspecific,
have a similar basis as role-related asymmetries. The reference of a
definite or specific noun phrase is established independently of that of a coargument
in a context-dependent way: the referent is identified by contextual
information or by the mutual knowledge of the speaker and hearer. By contrast,
the interpretation of an indefinite, non-specific noun phrase is determined by a
local binder that is structurally more prominent than it, i.e. c-commands it (cf.
Heim 1982). This means that the reference of an indefinite, non-specific argument
is determined locally by a structurally superior noun phrase. This explains why a
definite, specific, structurally superior subject and an indefinite, non-specific,
structurally inferior object are favoured in grammar and language processing. This
constellation is shown in (9):
(9) Every woman loves a man / the man.
In the indefinite, non-specific reading of the object (‘every woman loves some
man’), the referent of the object is determined by that of the subject. In this
reading, the number of men that are loved by at least one woman in the situation is
determined by the number of women that love at least one man in the situation.
However, in the definite or specific reading of the object (‘there is a particular
man that every woman loves’), the referent is established independently from that
of the subject. In this reading, the referents of the co-arguments are established
independently from one another. In sum, the connection between animacy-based
and reference-related DOM is that an animate or definite, specific object leads to a
departure from a uniform asymmetric co-argument dependency.
A generalized co-argument dependency notion that subsumes role-related and
reference-related dependencies is supported by neurolinguistic evidence
(Bornkessel-Schlesewsky and Schlesewsky 2009). Activation differences in the
left pSTS, which were discussed in connection with animate vs. inanimate objects
in (7a, b) above, are also manifest in manipulations of definiteness and specificity.
This effect occurs in German sentences with an indefinite, non-specific subject
and a definite (specific) object in comparison with sentences with a definite
(specific) subject and an indefinite, non-specific object (cf. also Kretzschmar et al,
this volume).

12
4. GENERALIZED SEMANTIC ROLES AND DOM
With the results of the last section in mind, let us return to DOM. The rolefunctional
view of DOM is formulated in (10) in terms of a licensing condition for
a differential object marker.
(10) Working hypothesis: A differential object marker is licensed by an object
whose intrinsic meaning properties qualifies it as a proto-agent in the
situation denoted by the predicate.
(10) is a working hypothesis as it is formulated too liberally to yield appropriate
descriptive results for individual languages. First, there are languages without
DOM, in which (10) does not operate: the intrinsic meaning of a noun phrase does
not influence case selection. Furthermore, languages may also vary with respect to
the nature of intrinsic properties that trigger DOM. DOM may be obligatory with
humans only, as in Malayalam, or with all higher animates, as in Spanish. An
explanation of the various patterns of variation is beyond the scope of this chapter.
For convenience, I assume that the licensing condition in (10) is grammaticalized
in various ways yielding patterns of DOM that need further explanatory steps.
Distinguishing meaning properties from world knowledge is a notoriously difficult
problem that is left aside here. Therefore, ‘meaning’ is used in a very broad sense
in (10).
On order to see how (10) works, let us take a closer look at DOM with
inanimate objects in Malayalam. In Malayalam, direct objects of transitive verbs
are obligatorily marked with a specific case, the suffix –e, if they are human. This
also holds for most, but not all animates. Definiteness is an additional factor that
influences the selection of the animate object marker. This distribution is similar
to that illustrated in section 2 above by examples from Spanish and Hindi. When
both arguments are inanimate in Malayalam, it is possible to mark the inanimate
object in order to resolve potential ambiguity, as shown in (11a, b).
Malayalam (Asher and Kumari 1997: 204)
(11) a. Kappal tiramaalakaÒ-e bheediccu.
ship waves-OBJ split.PST
‘(The) ship broke through (the) waves.’
b. TiramaalakaÒ kappal-ine bheediccu.
waves ship-OBJ split.PST
‘(The) waves split (the) ship.’
With the verb bheediccu ‘split’ in (11a, b), both the ship and the waves qualify as
proto-agents due to their intrinsic meaning: ship and wave refer to an entity with
its own source of energy that enables one to split the other. In Dowty’s terms, both
the ship and the waves have intrinsic meaning properties that qualify them as
autonomous movers and causers of the event denoted by the predicate. This

13
situation licenses a differential object marker, as stated in (10).
In the next pair of examples only the subject referent is capable of causing the
respective event. Cf. (12a, b) from Asher and Kumari (1997: 204):
(12) a. Tiiyyə kuˇil naSippicu.
fire hut destroy.PST
‘Fire destroyed (the) hut.’
b. VeÒÒam tiiyyə keˇutti.
water fire extinguish.PST
‘Water extinguished (the) fire.’
Part of the meaning of fire is that it may cause the destruction of a hut and part of
the meaning of water is that it may cause the extinction of a fire, but not vice
versa. In (12a, b), the interpretation of the subject as a proto-agent, specifically as
a causer, and the interpretation of the object as a proto-patient, i.e. as causally
affected, is unequivocal. In this event, the animate object marker is usually
omitted in Malayalam, particularly if the object is interpreted as indefinite. In sum,
the data in (11) and (12) support the view defended here that animacy-based DOM
is motivated by role-semantic interpretation, as stated in (10). Additionally, we
have to assume that Malayalam has incorporated (10) in its grammar in ways that
have to be specified in greater detail in order to capture further intricacies
pertaining to the optionality of the object marker and its reference-related
occurrence (p.c. Thomas Anzenhofer).
In the next step, we will take a closer look at DOM in Spanish. The data
in (13) – (15) are taken from García García (2007). Let us begin with the
straightforward case illustrated in (1) above as a typical specimen of DOM. It is
repeated in (13) for convenience:
(13) a. Conozco *este actor / a este actor.
know:PRS.1SG this:M.SG actor / OBJ this:M.SG actor
‘I know this actor.’
b. Conozco esta película / *a esta película.
know:PRS.1SG this:F.SG film / OBJ this:F.SG film
‘I know this film.’
My interpretation of this pattern is that an animate object but not an inanimate one
has intrinsic meaning properties that qualify it as a proto-agent in the situation
denoted by the predicate. Part of the meaning of actor ‘actor’ is that it denotes a
human being that is capable of knowing something or somebody. This is the agent
component of sentience. Thus, actor licenses DOM in (13a). In contrast, the
meaning of película ‘film’ does not allow an interpretation as an experiencer (or a
cognizer). Accordingly, DOM is not licensed in (13b).
The instances involving inanimate objects in Spanish can be explained in a

14
similar way. A remarkable minimal pair is offered in (14a, b):
(14) a. El profesor reemplaza el libro.
the professor replace.PRS.3SG the book
‘The professor replaces the book (with something else).’
b. El profesor reemplaza al libro.
the professor replace.PRS.3SG OBJ:DEF.M.SG book
‘The professor takes the place of the book.’
As noted by García García (2007), among others, there is a subtle, but
nevertheless crucial meaning difference between (14a) and (14b). The sentence
(14a) without a-marking means that the professor replaces the book with another
book or with something else, for instance, in a shelf. In this reading the book lacks
proto-agent properties, since the capability to replace things in the sense conveyed
by (14a) it is not part of the meaning of books. The sentence (14b) with a-
marking, however, means that the professor takes the place of the book: he fulfils
the didactic, entertaining or whatever function is attributed to the book in the
situation expressed by this sentence. Part of the meaning of books is that they
share with humans the capacity of inform, entertain or bore us. Thus, (14b)
implies that the professor and the book have the capability to act as proto-agents in
an equivalent way. Since reemplazar does not specify the exact nature of the
activity in which the professor gets involved instead of the book, we cannot
specify the proto-agent entailments with certainty. If, for example, the implicit
activity in (14b) is related to tuition, the professor causes that somebody gets to
know something and so does the book in other circumstances.
The example offered in (3) above and repeated here in (15) is similar:
(15) En esta receta, la leche puede sustituir al huevo.
in recipe the:F.SG milk can:PRS.3SG replace OBJ:DEF.M.SG egg
this:F.SG
‘In this recipe, egg can be replaced by milk.’
The interpretation of (15), where object a-marking is obligatorily required, is
similar to that in (14b): the subject participant and the object participant have the
capability to behave in an equivalent way in the situation denoted by the verb
sustituir. Since sustituir does not specify the exact nature of the event in which the
milk gets involved instead of the egg, we cannot specify the proto-agent
entailments out of context, but a plausible interpretation is that eggs and milk
cause a change in the food: it will contain more protein and fat. 6
6 Instead of causing a problem to a generalized approach to semantic roles (as suggested by a reviewer
of this chapter), underspecified verbs such as sustituir and reemplazar rather support this approach.
Strict entailments are that both participants have the capacity to exhibit an equivalent proto-agent
behaviour, with the difference that only the subject participant is actually involved instead of the object
participant. These entailments suffice to licence the a-marker on the object. Specifications regarding

15
The verbs reemplazar and sustituir, whose meanings entail that both
participants have the capacity to exhibit an equivalent proto-agent behaviour, are
typical for the class of verbs that are frequently used with an inanimate a-marked
object. Such verbs denote an interaction or a symmetrical situation (x is / acts like
y), for example, acompañar ‘accompany’, preceder ‘precede’, seguir ‘follow’,
and corresponder ‘correspond’. Verbs of classification and identification are
semantically similar and also occur with inanimate a-marked objects, e.g.
considerar ‘consider’, clasificar ‘classify’, caracterizar ‘characterize’, concretar
‘make specific’, designar ‘designate’, definir ‘define’, distinguir, diferenciar
‘distinguish’, especificar ‘specify’, and llamar ‘call, consider to be’ (cf. García
García 2007 for more details). 7
The data with inanimate objects taking the differential object marker
support the role-functional view defended here. It is not animacy per se that
counts but rather the semantic function of the object. It must be a potential protoagent
in the situation denoted by the predicate.
Previous studies have acknowledged the agentivity-related function of
DOM. Closest to the present approach is García García (2007). This study is based
on generalized semantic roles (e.g. Primus 1999). In his view, it is only DOM with
an inanimate object that is correlated primarily with the role-semantic relation
between subject and object: when the direct object is equally or more agentive
than the subject a-marking is required according to García García. My treatment
differs from his in some respects. It treats an animate and an inanimate object as
on a par and explains the distribution of the dative prepositional marker in all
types of constructions, as will be shown below.
In his cross-linguistic study, Kittilä (2006) also assumes that the function
of DOM is to distinguish between two potential agents. Regarding Spanish,
Delbecque (1998, 2002) assumes that the a-marker indicates in dynamic processes
that they involve potential agents as their objects. For Delbecque, these are
participants susceptible to trigger an action in reaction to their own way of being
or behaving. Kittilä’s and Delbecque’s approaches are very close to the approach
presented here but the scope of their explanation is unduly narrowed as a result of
using a traditional agent concept that is restricted to volitional processes. As a
consequence, DOM with stative verbs in Spanish, including conocer ‘know’,
illustrated in (1) and (13) above, as well as the high number of stative verbs
selecting the a-marker for inanimate objects cannot be explained. There is in fact
no language where DOM is restricted to verbs that subcategorize for volition.
DOM in ditransitive constructions with two animate objects also pose a
problem for approaches using the traditional agent notion. Such approaches lack
the theoretical means to treat a recipient as a proto-agent, a move that is necessary
in order to explain DOM in ditransitive clauses. By contrast, in my approach, a
the type of involvement are added by the context and are not part of the verb meaning in a narrow
sense.
7 Locative verbs select the a-marker according to the original locative meaning of this preposition and
fall beyond the scope of the present discussion, e.g. colocar ‘place’, poner ‘put’, situar ‘locate’, llegar
‘reach’.

16
recipient (or a recipient-like role) is characterized by a combination of proto-agent
and proto-patient components, as stated in the previous section. This explains why
in ditransitive constructions, two animate objects may receive the same marker
(cf. the Punjabi example in (5a) above). An additional explanation is needed for
the suppression of DOM in ditransitive clauses, as shown in the Spanish example
(4a) above. This may be due to a higher-ranking constraint that bans multiple
datives in Standard Spanish (as in some other languages). Under these
circumstances, two roles compete for the dative: the recipient and the animate
patient. In Standard Spanish, the competition is resolved in favour of the protoagent
components that are entailed by the meaning of the verb, i.e. subcategorized
for by the verb. This means that the dative is used for the recipient and blocked for
the animate patient (cf. also fn. 9 below).
The polyfunctionality of the animate object marker can be straightforwardly
explained in the present approach. As mentioned earlier in this chapter, in most of
the Indic, Iranian, Semitic, Tupi-Guarani, and Romance languages, the animate
object marker is also used for the recipient, addressee, and benefactive (i.e. the
dative) in ditransitive constructions. In some Romance, Iranian and Indic
languages it also codes the experiencer of certain psychological predicates. In
some Iranian languages (e.g. Jaghnobi) it also indicates the agent of transitive
clauses (i.e. the ergative). In Malayalam the object marker is used, besides the
dative and comitative, with the addressee or recipient of a few ditransitive verbs,
e.g. pahippiccu ‘taught’ and eelpiccu ‘entrusted’ (cf. Asher and Kumari 1997:
108). All these roles include agentive components and all the above-mentioned
cases, specifically the dative and the ergative, are regularly linked to roles with
agentive components.
5. AN OPTIMALITY-THEORETICAL TREATMENT OF
DOM IN SPANISH
In order to demonstrate the features of the present approach in more precise terms,
I will offer an optimality-theoretical treatment of DOM in Spanish. Optimality
Theory (OT) views grammar as an optimization procedure. For a given input the
grammar generates a set of output candidates by means of a generator. These
candidates are evaluated with respect to a set of constraints, which are violable,
potentially conflicting, and ranked with respect to one another. The output
candidate with the best constraint violation profile is the optimal candidate. OT is
a convenient method of capturing the competition between functional
transparency, i.e. faithfulness to the role-semantic input in our discussion, and
formal economy, specifically avoidance of overt case marking. A faithful
expression of meaning (BE PRECISE) is optimal for the hearer, formal economy (BE
BRIEF) is optimal for the speaker.
In the DOM patterns under discussion, overt case marking of the object is
restricted to situations in which the meaning of the object is marked. Semantically

17
unmarked objects are linked to patients, but in DOM, the patient referent has also
properties that qualify it for the agent role. Assuming that objects with mixed role
properties have marked meanings, the DOM patterns under discussion are
instances of constructional iconism. Constructional iconism means that unmarked
forms have unmarked meanings, and conversely, that marked forms are associated
with marked meanings.
In standard OT, syntax and semantics are disconnected. The independent
application of syntax and semantics does not yield a model that assigns the
consistent relation between form and meaning that is found in constructional
iconism. Bidirectional OT, an extension of regular (unidirectional) OT, is an
appropriate method to capture the form-meaning relation found in constructional
iconism. 8 In bidirectional OT, the input and the candidates consist of formmeaning
pairs . For our discussion we need case-role pairs. There are
different possibilities for comparing pairs of form and meaning. The one we need
for the Spanish data (and for variation due to constructional iconism in general) is
weak bidirectional OT. In this model, the notion of superoptimality plays an
important role and is defined as follows (cf. Jäger 2002, Krifka 2002):
(16) A pair of a set of candidates is superoptimal iff:
i. There is no superoptimal in the candidate set such that
is more optimal than .
ii. There is no superoptimal in the candidate set such that
is more optimal than .
The notion of superoptimal pairs is restricted to those pairs that have no
competitor on the expression level or on the meaning level that is itself
superoptimal. Pairs that do not share an identical form or meaning are not
competitors. As a result, more than one candidate may emerge as superoptimal.
The two competing violable constraints that are needed to capture DOM in
Spanish are given in (17) and (18):
(17) ECONOMY (BE BRIEF): Overt case is not used.
(18) EXPRESS ROLE (BE PRECISE): The dative is used for an argument with a
low number of proto-agent properties that are i) subcategorized by the
verb or ii) assigned according to the intrinsic meaning of the direct object
referent.
The economy constraint is explicitly or tacitly formulated in this or in a similar
way by many approaches to cases in various lines of research.
The dative constraint is taken from Primus (1999, chap. 4) where it is used for
various typological data. In (18) it is adapted to capture the dative as a differential
8 Another version of bidirectional OT has been successfully applied to DOM in de Swart (2007).
However, in this approach animacy and traditional semantic roles are taken as distinct factors for case
selection.

18
object marker along the lines of the more general working hypothesis in (10)
above. In DOM-languages using the dative as a differential object marker, the
dative is sensitive to the intrinsic meaning of the object referent, as discussed in
the previous section. This is achieved in (18) by taking both subcategorized
properties (condition (18i)) and intrinsic properties (condition (18ii)) into account.
In languages with DOM, condition (18ii) holds in addition to condition (18i),
which is the basic condition holding irrespective of DOM in all languages with an
agentive function of the dative. 9 This in accordance with the assumption that
subcategorization is a universal condition for case selection. The restriction to a
low number of proto-agent properties is meant to exclude dative volitional agents.
Volitional agents must surface as nominative subjects in the basic (active)
construction in Spanish as in many other accusative languages.
If the dative preposition a is selected for subcategorized roles in Spanish, the
respective argument is usually cross-referenced by the dative clitic (e.g. le for
masculine, singular, as in the examples in this paragraph). In DOM, the dativemarked
argument is usually taken up by an accusative clitic (lo for masculine,
singular). The verbs that select the dative in Spanish are very similar to the dative
verbs in other languages (e.g. German). These are experiencer verbs (e.g. al
artista le gusta la cerveza ‘the artist (DAT) likes the beer’), verbs of alienable
possession (e.g a Pedro le pertenece esta casa ‘this house belongs to Peter’),
verbs denoting an involuntary action (e.g. a Pedro se le ha roto el vaso ‘Pedro
(DAT) broke the glass inadvertently’), verbs denoting an interaction or a
symmetrical situation such as ayudar ‘help’, seguir ‘follow’, and parecer
‘resemble’, and last but not least, ditransitive verbs selecting a recipient, an
addressee or a benefactive (cf. the examples given in (4a, b) earlier in this
chapter).
For illustrative purposes I restrict the number of OT candidates to those
that have a chance to win the competition imposed by the two constraints under
discussion. I take the options for conocer and reemplazar, see (13) – (14) above,
as illustrations. The potential winners fall into four form-meaning pairs, which are
discussed in turn.
In the first type of candidates, the object lacks overt case and has no intrinsic
agentive property. This form-meaning pair is abbreviated as . The
candidates are conozco esta película and el profesor reemplaza el libro in the
patient reading ‘the professor replaces the book with something else’. They
perfom best relative to their competitors with respect to ECONOMY and EXPRESS
ROLE. They have no overt case and fulfill the dative constraint vacuously since the
object lacks agentive properties. This type emerges as superoptimal.
The second type is characterized as follows: the object bears dative marking
and has no intrinsic agentive property. This type is abbreviated as . The
candidates are conozco a esta película and el profesor reemplaza al libro in the
9 The interaction between the subcategorization condition (18i) and the DOM-condition (18ii) can be
captured in OT-terms, in principle. In case of conflict, (18i) is higher ranked than (18ii) in Spanish.
This is illustrated by the ditransitive construction (4a, b) in section 2 of this chapter. In this
construction, the animate patient looses is marker in favour of the subcategorized recipient.

19
patient reading ‘the professor replaces the book with something else’. These
candidates violate both ECONOMY and EXPRESS ROLE, i.e. the dative constraint,
and are eliminated. This happens because they loose against a legitimate
competitor. i.e. , that has the same meaning but incurs no violation.
They also loose against the fourth type of candidate, i.e. , which has the
same case but a better meaning for this case.
In the third type, the object lacks overt case and has intrinsic agentive
properties. This type is abbreviated as . The candidates are conozco
este actor and el profesor reemplaza el libro in the agentive reading ‘the professor
takes the place of the book’. These candidates violate EXPRESS ROLE, i.e. the
dative constraint, and are eliminated. This happens because they loose against
their legitimate competitors. has the same form but a better meaning
for this form; has the same meaning but a better form for this meaning.
Finally, in the fourth type, the object bears dative marking and has intrinsic
agentive properties. This type is abbreviated as . The candidates are
conozco a este actor and el profesor reemplaza al libro in the agentive reading
‘the professor takes the place of the book’. These candidates violate ECONOMY,
but fulfill EXPRESS ROLE, i.e. the dative constraint. They loose only relative to
candidates that are not winners, so that turns out to be superoptimal as
well.
The tableau in (19) illustrates the evaluation procedure for the input
conocer(x,y) that entails SENTIENCE(x,y). The arguments x and y are specified in
the input as follows: x is the first person singular (the speaker) and y is esta
película or este actor. The evaluation considers both ranking options of the two
constraints under discussion since their relative ranking is not crucial. This is
indicated by the dotted line. The winner is the candidate that has the smallest
number of violations of the relevant highest constraint. A hand () points to the
winner. The plus sign (+) indicates the candidate that does not compete with the
winner. Recall that pairs that do not share an identical form or meaning are not
competitors in this model. According to the logic of weak bidirectional OT, the
winner and the non-competitor come out as superoptimal.
(19) Evaluation for the input conocer(x,y); x = 1SG; y = esta película or este
actor
ECONOMY EXPRESS ROLE
1 conozco esta película
2 conozco a esta película * *
3 conozco este actor *
4 +conozco a este actor *
would be the only winner in unidirectional OT. In weak bidirectional
OT is also a winner (i.e. a superoptimal candidate) because there is no
direct competitor that is superoptimal. does not compete with the
superoptimal candidate because it has a different form and a different
meaning. Since these two candidates do not compete, they emerge as superoptimal
in this model. This result is independent of the assumed ranking. The bidirectional

20
model captures the fact that a marked form (i.e. the dative) may be optimal
provided it expresses a marked meaning (i.e. a patient with agentive properties).
The OT treatment highlights the claims that characterize the present
approach in more explicit terms. Case selection based on animacy (DOM) and role
semantics are captured by one faithfulness constraint in Spanish. This explains the
polyfunctionality of the differential object marker. Semantic subcategorization, i.e.
semantic-role information that is anchored in the meaning of the predicate, is the
basic condition for case-selection. In DOM-languages such as Spanish, intrinsic
meaning properties of the object referent are interpreted in terms of semantic roles
and influence case selection as an additional condition. The close connection
between subcategorized and intrinsic properties also explains why the semantic
verb class may influence DOM. The appeal of the present proposal is that the
constraints are not restricted to DOM, but needed on independent grounds. The
present proposal incorporates most of the explanations of DOM offered in the
literature but captures more data in more precise terms.
First, as already mentioned above, it accommodates the claim of previous
proposals that the function of DOM is to mark potential agents (cf. Kittilä 2006,
Delbecque 1998, 2002). As mentioned, the scope of these approaches is unduly
restricted to traditional agents and to verbs that select an agent.
Furthermore, the present proposal incorporates the markedness view on DOM
(cf. Comrie 1989, Aissen 2003). It is formulated by Comrie (1989: 128) in terms
of the grammatical functions used in typological research – A for transitive subject
and P for object – as follows: “In a standard transitive scenario, A is animate and
definite while P is inanimate/ indefinite (or at least less animate and definite than
A), so any deviation from this scenario should be marked”. This account correctly
predicts that animate objects should be case-marked, while inanimate objects need
not be marked. The advantage of the present approach is that it explains the
correlations of the standard transitive scenario. Roles that are canonically linked to
A fall under proto-agent and entail animacy in most cases, whereas roles
canonically reserved for P are proto-patients that lack animacy entailments. 10 A
further appeal of the present approach is that it also explains the standard
ditransitive scenario, in which recipients are also preferably animate. Their bias
towards animacy results from the fact that they have proto-agent properties that
entail animacy, e.g. alienable possession or sentience. Taking recipients into
account is particularly useful in view of the fact that many animate object markers
are also recipient markers.
Finally, there are recent approaches that explain DOM by the
distinguishability and characterizing function of cases (e.g. de Hoop and Lamers
2006, de Swart 2007, Malchukov 2008). The two functions of cases are described
10 Corpus studies support the correlations between animacy and grammatical or semantic functions.
Subjects of transitive clauses (A), which bear an agentive role, are predominantly animate. This
percentage varies according to the language or the language variety under investigation, for example,
from 69% in Norwegian (Øvrelid 2004) to 93% in spoken Swedish (Dahl 2000). In contrast, objects,
which bear a patient-like role, refer predominantly to inanimate entities (89% of the direct objects in
Dahl’s corpus, 90% in Øvrelid’s sample).

21
in the typological literature as follows (cf. Comrie 1989: 117f., Song 2001: 156f.):
In their distinguishing function, cases are used for discriminating concomitant
constituents with different semantic or syntactic properties. This function explains
the fact that the subject of a one-place predicate is generally marked by the same
case, the nominative or absolutive, irrespective of its semantic role. Ideally, a
second case is only used with predicates selecting at least two roles and a third
case is only required by predicates with at least three roles. The second broad
function of cases is the characterising one. Ideally, a case has this function if it is
used for all and only the noun phrases with a certain type of semantic role or
syntactic function. The explanation of DOM in terms of distinguishability is that
two animate participants have to be distinguished formally. A distinction is not
needed if agents and patients are distinguished by animacy. The present approach
captures both functions of cases by combining economy and role faithfulness
constraints. As shown by de Swart (2007) and Malchukov (2008), these two
factors do not compete in DOM-patterns. This is also the result of the present
treatment. Recall that the DOM-pattern emerges as a winner, irrespective of the
ranking of the two constraints proposed in (17) and (18). The advantage of the
present approach is that it pins down the characterizing function in more precise
terms. It suffices to repeat here that the dative constraint explains subcategorized
datives as well as datives used for animate patient objects in Spanish.
6. EXPLAINING THE INTERACTION BETWEEN
ANIMACY AND SEMANTIC ROLES
The previous sections were focussed on the claim that animacy and role semantics
are closely connected. They have presented grammatical and processing evidence
for this claim as well as a formal treatment in optimality-theoretic terms. This last
section will address the deeper question regarding the nature of this connection.
Given the fact that animacy is unilaterally implied by most of the agentive
basic notions, such as volitional causation, sentience, alienable possession, and
certain patterns of autonomous movement, I still need to explain why the
inference is reversed from animacy to agentivity. A plausible explanation is
abductive reasoning (cf. Levinson 2000). This is a standard type of pragmatic
inference by which a unilateral implication or conditional is reversed. As a result,
a unilateral implication is pragmatically strengthened to a bilateral implication
thereby increasing the informative content of an utterance. This tendency to
increase the informativity of a linguistic expression explains why animacy is used
as a cue for agentivity in language processing.
The pragmatic inference from animacy to agentivity has not gone unnoticed
in role-semantic approaches. An intriguing aspect of agentivity is that volition or
control is not subcategorized for as often as expected from its prime status in role
semantics. Take for example the verbs in (20):

22
(20) a. John / the stone rolled down the hill.
b. This medicine / Jogging / Rebecca helped Jamaal.
The most straightforward analysis of the meaning of roll and help is that these
verbs are semantically underspecified for volition. This captures the range of uses
illustrated in (20). As assumed by Foley and Van Valin (1984) and Engelberg
(2005), among others, volition is assigned by default interpretation if the subject
argument is animate (or human).
The pragmatic explanation is in compliance with the principle of incremental
language processing, which is adopted by many neurolinguistic investigations (cf.
Friederici 1999, Schlesewsky and Bornkessel 2004, de Hoop and Lamers 2006
with a focus on language comprehension, and Branigan et al. 2008 for language
production). According to this principle, linguistic information is parsed as soon
as possible, thereby maximizing processing efficiency. Incremental processing is
maximized under the assumption that language users make probabilistic syntactic
choices based on violable principles of grammar and multidimensional
information (Bresnan et al. 2001). Turning to our topic, this means that there is an
advantage in making role-semantic interpretation immediately available by using
grammatical principles and any type of relevant information, including case, word
order, verb agreement, and animacy (cf. de Hoop and Lamers 2006).
Harmonically aligned asymmetries pertaining to semantic roles, cases, word order,
and animacy maximize processing efficiency. Disharmonic alignments are costly
(cf. Bornkessel-Schlesewsky and Schlesewsky 2009, Lamers, this volume). This
happens, for instance, when semantic-role prominence (e.g. agent > patient) does
not match animacy prominence (e.g. animate > inanimate).
The link that is still missing in the previous argumentation is provided by the
assumption that grammar is based on processing preferences to a considerable
degree. This assumption is shared by many recent approaches of different
provenience (cf. Hawkins 1994, Bybee and Hopper 2001, Bresnan et al. 2001,
Culicover and Jackendoff 2005, Newmeyer 2005). In this view, grammatical
structures emerge in proportion to their preference in processing. This also means
that soft constraints, i.e. tendencies or preferences, in one language may be
inviolable hard constraints in another language (Bresnan et al. 2001). Taking a
pertinent example, in many languages, a second noun phrase is interpreted as a
patient object per default. If this noun phrase refers to an animate being, its
intrinsic meaning is incongruous with its default interpretation as a patient. This is
a situation that is dispreferred in processing, as mentioned in section 2 above. In
OT terms (cf. de Hoop and Lamers 2006), animacy is a low-ranked, soft constraint
that only kicks in, if case and word order do not disambiguate semantic roles.
However, in languages with DOM, such as Hindi and Spanish, there are
grammatical means to express the dispreferred correlation.
So far, I have pinned down the mechanisms by which animacy, an intrinsic
property of a noun phrase referent, is interpreted in terms of agentivity. But I have
not yet addressed the question of why agentivity implies animacy. This question
leads us to the nature of semantic roles. According to Dowty (1991: 561), protorole
features are those semantic features that are relevant for argument (subject or

23
object) selection. This is a good criterion but still leaves one fundamental question
unanswered. Why is argument selection sensitive to notions such as volition,
motion and sentience, and not to basic geometric features (e.g. being a circle) or
basic physical notions (e.g. being fluid)?
A deeper insight into the nature of semantic roles can be extracted from
Dahl’s and Fraurud’s assumptions about the nature of animacy (cf. Dahl and
Fraurud 1996, Dahl 2008). In their view, animacy is at the bottom a question of
the distinction between ‘‘persons, that is, essentially human beings perceived as
agents, and the rest of the universe [...] Indeed, the notion of ‘‘personhood’’ seems
to embody what is quintessential to animate beings, both the roles as agent and
experiencer, and the focus on the individual” (Dahl 2008: 145-146). This view
supports the present assumption that animacy and agentivity are closely
connected.
The additional appeal of this type of explanation is that it ties animacy to
individuation, i.e. reference types (Dahl 2008: 149). Dahl assumes that animates
are easier to grasp as individuals than inanimates. Within the class of animates the
self and other individuals who can perceive the world and act upon it as myself are
central. For Dahl, this means that we use ourselves as models for others. In this
way, we obtain a finer grained cognitive scale, corresponding to a more elaborate
nominal hierarchy: the self with his / her properties (e.g. proto-agent, human, 1 st
person, definite, singular, countable, etc.) is the model for others. The position of a
category on this scale is determined by the grade of its similarity to the self, i.e. by
the number of properties it shares with the self. This is a promising first step
towards a better understanding of the interaction between reference types and
animacy in grammar and language processing.
If Dahl’s claim is correct that animacy is a phylogenetically deeply anchored,
fundamental (ontological) category, then proto-agent properties are arguably
dependent on individuation and animacy. I therefore hypothesize that proto-agent
properties 11 can be explained as those cognitive-semantic properties that
characterize how a human individual, in particular oneself, is involved in
situations.
Dahl’s (2008: 145-146) claim that ANIMATE and INANIMATE are ontological
types (in the strict philosophical sense) has another interesting consequence. For
Dahl, one feature of ontological types is the common lack of natural ways of
referring to them. For instance, English seems to have no generic noun for
‘animate being’. Even words for ‘human’ tend to be identical to or derived from
words meaning ‘male being’. Rather, ontological types work behind the scene,
channeling the ways we speak about entities in the world. In my view, Dahl’s
assumptions explain why animacy per se is not crucial in grammars, as argued in
this chapter. Instead, manifestations of animacy, particularly proto-agentivity and
individuation, are relevant in language.
11 Recall that proto-patient properties are derivative in my approach and involve the same basic
notions that are needed to define proto-agent.

24
REFERENCES
Ackerman, Farrell, and John Moore. 2001. Proto-properties and grammatical encoding: A
correspondence theory of argument selection. Stanford: CSLI Publications.
Aissen, Judith. 2003. Differential object marking: Iconicity vs. economy. Natural Language and
Linguistic Theory 21: 435–483.
Asher, Ronald E., and T. C. Kumari. 1997. Malayalam. London/New York: Routledge.
Bhatia, Tej K. 1993. Punjabi. London: Routledge.
Bornkessel-Schlesewsky, Ina, and Matthias Schlesewsky. 2009. The role of prominence
information in real time comprehension of transitive constructions: A cross-linguistic
approach. Language and Linguistics Compass 3, 19-58.
Bossong, Georg. 1985. Differentielle Objektmarkierung in den Neuiranischen Sprachen.
Tübingen: Narr.
Bossong, Georg. 1998. Le marquage différentiel de l’object dans les langues d'Europe. In
Actance et valence dans les langues de l'Europe, ed. Jack Feuillet, 193–258. Berlin: de
Gruyter.
Branigan, Holly, Martin J. Pickering, and Mikihiro Tanaka. 2008. Contributions of animacy to
grammatical function assignment and word order during production. Lingua 118: 172–189.
Bresnan, Joan, Shipra Dingare, and Christopher D. Manning. 2001. Soft constraints mirror hard
constraints: Voice and person in English and Lummi. Proceedings of the Lexical Functional
Grammar Conference 2001. Stanford: CSLI Publications.
Bybee, Joan, and Paul Hopper (eds.). 2001. Frequency and the emergence of linguistic structure.
Amsterdam: John Benjamins.
Comrie, Bernard. 1989. Language universals and linguistic typology. 2 nd ed. Cambridge:
Blackwell.
Culicover, Peter W., and Ray Jackendoff. 2005. Simpler syntax. Oxford: Oxford University
Press.
Dahl, Östen, and Kari Fraurud. 1996. Animacy in grammar and discourse. In Reference and
referent accessibility, eds. Thorstein Fretheim and Jeanette K. Gundel, 47–64. Amsterdam:
John Benjamins.
Dahl, Östen. 2008. Animacy and egophoricity: Grammar, ontology and phylogeny. Lingua 118,
141–150.
de Hoop, Helen, and Monique Lamers. 2006. Incremental distinguishability of subject and
object. In Case, valency and transitivity, eds. Leonid Kulikov, Andrej Malchukov, and Peter
de Swart, 269–287. Amsterdam / Philadelphia: John Benjamins.
de Swart, Peter, Monique Lamers, and Sander Lestrade. 2008. Animacy, argument structure, and
argument encoding. Lingua 118: 131–140.
de Swart, Peter. 2007. Cross-linguistic variation in object marking. Utrecht: LOT publications.
Delbecque, Nicole. 1998. Why Spanish has two transitive construction frames. Leuvense
Bijdragen 87: 387–415.
Delbecque, Nicole. 2002. A construction grammar approach to transitivity in Spanish. In The
nominative / accusative and their counterpart, eds. Kirstin Davidse and Beatrice Lamiroy,
81–130. Amsterdam: John Benjamins.
Dixon, Robert M. W. 1994. Ergativity. Cambridge: Cambridge University Press.
Dowty, David R. 1991. Thematic proto-roles and argument selection. Language 67: 547–619.
Engelberg, Stefan. 2005. Stativity, supervenience, and sentential subjects. In Event arguments.
Foundations and applications, eds. Claudia Maienborn and Angelika Wöllstein, 45–68.
Tübingen: Niemeyer.
Fillmore, Charles J. 1968. The case for case. In Universals in linguistic theory, eds. Emmon
Bach and Robert T. Harms, 1–90. New York: Holt, Rinehart & Winston.

Fillmore, Charles J. 1977. The case for case reopened. In Grammatical relations, eds. Peter Cole
and Jerrold Sadock, 59–82. New York: Academic Press.
Foley, William A., and Robert D. Van Valin. 1984. Functional syntax and universal grammar.
Cambridge: Cambridge University Press.
Folli, Raffaella, and Heidi Harley. 2008. Teleology and animacy in external arguments. Lingua
118: 190–202.
Friederici, Angela D. 1999. The neurobiology of language processing. In Language
comprehension: A biological perspective, ed. Angela Friederici, 265-304. Berlin: Springer.
Frisch, Stefan, and Matthias Schlesewsky. 2001. The N400 indicates problems of thematic
hierarchizing. NeuroReport 12: 3391–3394.
García García, Marco. 2007. Differential object marking with inanimate objects. In Proceedings
of the workshop “Definiteness, specificity and animacy in Ibero-Romance languages, eds.
Georg A. Kaiser and Manuel Leonetti, 63–84. Working paper, University of Konstanz.
Gelman, Rochel, Frank Durgin, and Lisa Kaufman. 1995. Distinguishing between animates and
inanimates: not by motion alone. In Causal cognition. A multidisciplinary debate, eds. Dan
Sperber, David Premack, and Ann J. Premack, 150–184. Oxford: Clarendon Press.
Grewe, Tanja, Ina Bornkessel, Stefan Zysset, Richard Wiese, Yves D. von Cramon, and Matthias
Schlesewsky. 2005. The emergence of the unmarked: A new perspective on the languagespecific
function of Broca’s area. Human Brain Mapping 26: 178–190.
Grewe, Tanja, Ina Bornkessel, Stefan Zysset, Richard Wiese, Yves D. von Cramon, and Matthias
Schlesewsky. 2006. Linguistic prominence and Broca’s area: The influence of animacy as a
linearization principle. NeuroImage 32: 1395–1402.
Grewe, Tanja, Ina Bornkessel, Stefan Zysset, Richard Wiese, Yves D. von Cramon, and Matthias
Schlesewsky. 2007. The role of the posterior superior temporal sulcus in the processing of
unmarked transitivity. NeuroImage 35: 343–352.
Grewe, Tanja. 2006. The neuronal reality of the nominal hierarchy: fMRI observations on
animacy in sentence comprehension. Dissertation, University of Marburg.
Gross, James J. (ed.). 2007. Handbook of emotion regulation. New York: Guilford Press.
Hawkins, John A. 1994. A performance theory of order and constituency. Cambridge: University
Press.
Heim, Irene. 1982. The semantics of definite and indefinite noun phrases. PhD. dissertation,
University of Massachusetts at Amherst.
Heusinger, Klaus von. 2008. Verbal semantics and the diachronic development of differential
object marking in Spanish. Probus 20: 1-31.
Jackendoff, Ray. 1990. Semantic structures. Cambridge, Mass.: MIT Press.
Jäger, Gerhard. 2002. Some notes on the formal properties of bidirectional optimality theory.
Journal of Logic, Language and Information 11: 427–451.
Kittilä, Seppo. 2006. The woman showed the baby to her sister: on resolving animacy-driven
ambiguity in ditransitives. In Case, valency and transitivity. eds. Leonid Kulikov, Andrej
Malchukov, and Peter de Swart, 291–308. Amsterdam / Philadelphia: John Benjamins.
Kretzschmar, Franziska, Ina Bornkessel-Schlesewsky, Adrian Staub, Dietmar Roehm, and
Matthias Schlesewsky. Prominence Facilitates Ambiguity Resolution: On the Interaction
between Referentiality, Thematic Roles and Word Order in Syntactic Reanalysis, this
volume.
Krifka, Manfred. 2002. Be brief and vague! And how bidirectional optimality theory allows for
verbosity and precision. In Sounds and systems: studies in the structure and change. eds.
David Restle and Dietmar Zaefferer, 439–458. Berlin: de Gruyter.
Lakoff, George. 1977. Linguistic gestalts. Papers from the 13th Regional Meeting of the Chicago
Linguistic Society 13: 236–287.
Lamers, Monique J.A. Argument linearization in Dutch: a multi-factorial approach, this volume.
Lazard, Gilbert. 2001. Le marquage différentiel de l'objet. In Language typology and language
universals, 2 nd vol., eds. Martin Haspelmath et al., 873–885. Berlin: de Gruyter.
25

26
Leslie, Alan M. 1995. A theory of agency. In Causal cognition. A multidisciplinary debate, eds.
Dan Sperber, David Premack, and Ann J. Premack, 121–141. Oxford: Clarendon Press.
Levinson, Stephen C. 2000. Presumptive meanings. The theory of generalized conversational
implicature. Cambridge: MIT Press.
Malchukov, Andrej. 2008. Animacy and asymmetries in differential case marking. Lingua 118:
203–221.
Mohanan, Tara. 1994. Argument structure in Hindi. Stanford: CSLI.
Newmeyer, Frederick J. 2005. Possible and probable languages: A generative perspective on
linguistic typology. Oxford: Oxford University Press.
Ochsner, Kevin N., and James J. Gross 2005. The cognitive control of emotion. Trends in
Cognitive Science 9: 242–249.
Opfer, John. E. 2002. Identifying living and sentient kinds from dynamic information: the case of
goal-directed versus aimless autonomous movement in conceptual change. Cognition 86: 97–
122.
Øvrelid, Lilja. 2004. Disambiguation of grammatical functions in Norwegian: Modeling
variation in word order interpretations conditioned by animacy and definiteness. In
Proceedings of the 20 th Scandinavian Conference of Linguistics, ed. Fred Karlsson.
University of Helsinki: Department of General Linguistics.
Pelphrey, Kevin A., James P. Morris, and Gregory McCarthy. 2004. Grasping the intentions of
others: The perceived intentionality of an action influences activity in the superior temporal
sulcus during social perception. Journal of Cognitive Neuroscience 16: 1706–1716.
Premack, David, and Ann J. Premack. 1995. Intention as psychological cause. In Causal
cognition. A multidisciplinary debate, eds. Dan Sperber, David Premack, and Ann J.
Premack, 185–199. Oxford: Clarendon Press.
Primus, Beatrice. 1999. Cases and thematic roles – Ergative, accusative and active. Tübingen:
Niemeyer.
Primus, Beatrice. 2006. Mismatches in semantic role hierarchies and the dimensions of role
semantics. In Semantic role universals and argument linking. Theoretical, typological and
psycholinguistic perspectives, eds. Ina Bornkessel, Matthias Schlesewsky, Bernard Comrie,
and Angela D. Friederici, 53–89. Berlin: de Gruyter.
Schlesewsky, Matthias, and Ina Bornkessel. 2004. On incremental interpretation: Degrees of
meaning accessed during sentence comprehension. Lingua 114: 1213–1234.
Scholl, Brian J., and Patrice D Tremoulet. 2000. Perceptual causality and animacy. Trends in
Cognitive Sciences 4: 299–309.
Schultz, Johannes K., Hiroshi Imamizu, Mitsuo Kawato, and Chris D. Frith. 2004. Activation of
the human superior temporal gyrus during observation of goal attribution by intentional
objects. Journal of Cognitive Neuroscience 16: 1695–1705.
Schultz, Johannes K., Karl J. Friston, John O’Doherty, Daniel M. Wolpert, and Chris D. Frith.
2005. Activation in posterior superior temporal sulcus parallels parameter inducing the
percept of animacy. Neuron 45: 625–635.
Song, Jae Jung. 2001. Linguistic typology: Morphology and syntax. Harlow and London: Pearson
Education.
Thompson, James C., Michele Clarke, Tennille Stewart, and Aina Puce. 2005. Configural
processing of biological motion in human superior temporal sulcus. Journal of Neuroscience
25: 9059–9066.
Tremoulet, Patrice D., and Jacob Feldman. 2000. Perception of animacy from the motion of a
single object. Perception 29 : 943–951.
Van Valin, Robert D., and LaPolla, Randy. 1997. Syntax. Structure, meaning and function.
Cambridge: Cambridge University Press.
Wang, Luming, Matthias Schlesewsky, Markus Philipp, and Ina Bornkessel-Schlesewsky. The
role of animacy in online interpretation in Mandarin Chinese, this volume.
Wright, Georg Henrik von. 1963. Norm and action. London: Routledge & Kegan.